474 



Fishery Bulletin 105(4) 



fish size or age in this analysis. Relative fishing effort 

 at the spatial scale of PMFC areas was not available 

 for these years. 



Initial examination of the data (for a qualitative as- 

 sessment) was performed for tagging projects conducted 

 in Puget Sound and the Strait of Georgia (24,408 tags 

 released and 5756 recovered). Many of these experi- 

 ments were conducted with varied goals (other than 

 that of estimating movement rates) and an attempt to 

 recover tags from distant areas was not undertaken. 

 Therefore, these data were excluded from the mark- 

 recapture model but were used to structure the spatial 

 extent of the analysis. 



Movement hypotheses 



Many researchers have noted seasonal changes in catch 

 rates in specific areas, and the temporally transient 

 appearance of aggregations of flatfish. For English sole, 

 these aggregations seem to be associated with the winter 

 spawning season (Alverson, 1960). Spawning of English 

 sole occurs from early fall through late spring, and most 

 growth occurs during the rest of the year. December and 

 April appear to be the first and last months of strong 

 spawning activity across all latitudes for English sole 

 (Castillo, 1995); migration associated with movement to 

 and from the spawning grounds could therefore reason- 

 ably be expected to take place in the fall and spring just 

 before and following this spawning activity. 



For all modeled tag recoveries, movement was re- 

 stricted to adjacent areas (north or south), and the 

 same rates of movement were applied to all areas along 

 the coast (Puget Sound and the Strait of Georgia were 

 excluded). This simplification restricted the P matrices 

 (areaxarea for each month) to nonzero values in only the 

 first off-diagonals, and repeated the same parameters 

 within each diagonal (Table 2). The time-increments 

 considered reflect the trade-off between biologically 

 realistic hypotheses and the likely constraints on com- 

 plexity in future stock assessments. English sole move- 



ment over large distances has been observed to occur 

 at a rate of three to eight kilometers per day (Forrester, 

 1969). With an average latitudinal span of 138 kilome- 

 ters per area, it could therefore take 17 to 46 days for 

 at least some English sole to cross a single area; this 

 rate of movement indicated that a one-month time- 

 step would be appropriate to accommodate interarea 

 migration. In the simplest hypotheses, movement was 

 considered to take place only at one time per year, and 

 all P matrices contained only zeros except for the month 

 in which movement occurred but allowed northerly and 

 southerly movement to differ. These models included 

 movement occurring in January (the standard break 

 for assessment years), October, November, May, and 

 June. An additional four hypotheses were considered 

 that included movement in the fall and spring and that 

 resulted in two different P matrices — one movement 

 applied in a spring month and one in the fall with two 

 unique parameters (a north and south movement) in 

 each. Five hypotheses included movement during more 

 than two months of the calendar year (Table 3). 



Bayesian implementation 



Prior distributions are required for each of the model 

 parameters (Table 4). Priors were selected to be non- 

 informative, allowing the likelihood function to domi- 

 nate the posterior probability distribution. However, the 

 choice of appropriate noninformative priors is difficult 

 and case specific, requiring estimation on the appropri- 

 ate scale for each parameter, often either diffuse (near 

 uniform) or uniform over log-space (Gelman et al., 1995). 

 In some cases, additional information was available with 

 which to constrain the priors. All movement parameters 

 were bounded between zero and one, and had uniform 

 prior density. The maximum possible reporting rate 

 was estimated by calculating one minus the proportion 

 of recovered tags with incomplete information (no area, 

 date, or both, and were not included in the analysis); 

 this value was 0.92 per tag lifetime. If tag loss rate was 



